Time-frequency representations of power

Table of Contents
Introduction Preparation Time-frequency analysis of slower frequencies (<30 hz) Plotting the TFR of the low frequency results Question 1 Question 2 Question 3 Question 4 Comparing attend right versus attend left Time-frequency analysis of faster frequencies (>30 hz) Plotting the TFR of the high frequency results Question 5 Question 6 Question 7 Question 8 Question 9 Publications and preregistration References

Introduction

Time-frequency representations of power are used to quantify brain responses in the frequency domain. These reposes are often interpreted to reflect task-related modulations of brain oscillations. In this section we will show how these response can be estimated by an FFT approach using a sliding time-window.

Preparation

Make sure FieldTrip is installed and path is set. Then run ft_defaults (only needs running once during each Matlab session).
ft_defaults
Set the path and the file names of the fif files. We will be using the MaxFiltered data.
data_path = 'C:\YOUR_DATA_FOLDER\';
Load the cleaned planar gradiometer data.
data=[data_path,'planar_clean.mat'];
load(data)
Select 'attend left' trials and 'attend right' trials for the two conditions.
idx=(find(planar_clean.trialinfo(:,1)==21))';
cfg=[];
cfg.trials=idx;
data_left=ft_selectdata(cfg,planar_clean);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds
clear idx
idx=(find(planar_clean.trialinfo(:,1)==22))';
cfg=[];
cfg.trials=idx;
data_right=ft_selectdata(cfg,planar_clean);
the call to "ft_selectdata" took 0 seconds

Time-frequency analysis of slower frequencies (<30 hz)

From experience we know that oscillations in slower frequency bands such as in the theta, alpha and beta range are fairly narrow-band (ΔF ~ 2 Hz) whereas neuronal activity in the gamma band (30 - 100 hz) is typically more broad-band (ΔF ~ 5-10 hz). For this reason we advise the use of different settings for slower and faster frequencies.
The goal is to calculate the time-frequency power per trials. The power will subsequently be averaged over trials. To calculate the power, the Fourier transform is applied with a sliding time-window approach. The length of the time-window (cfg.t_fimwin) is ΔT = 0.5 s. It will be moved in time steps of 0.05 s starting at -1.8 and ending at 1 (cfg.toi). Prior to calculating the Fourier transform the time window is multiplied by a Hanning taper (also length ΔT). The analysis will be performed from 1 to 30 Hz in steps of Δf = 2 Hz (cfg.foi). Note that the frequency steps are dictated by the length of the sliding time-window: Δf = 1/ΔT. The spectra smoothing is about ΔFs = 1.5 / ΔT = ±3 Hz (the factor 1.5 stems for the Hanning taper).
Loc_eps_stft_spec.gif
Calculate the time-frequency power for both conditions:
cfg = [];
cfg.output = 'pow';
cfg.channel = 'MEGGRAD';
cfg.taper = 'hanning';
cfg.method = 'mtmconvol';
cfg.foi = 2:2:30;
numfoi = length(cfg.foi);
cfg.t_ftimwin = ones(length(cfg.foi),1).* 0.5;
cfg.toi = [-1.8 : 0.05: 1];
cfg.keeptrials = 'no';
tfr_low_left = ft_freqanalysis(cfg, data_left);
the input is raw data with 204 channels and 128 trials the call to "ft_selectdata" took 0 seconds Default cfg.pad='maxperlen' can run slowly. Consider using cfg.pad='nextpow2' for more efficient FFT computation. processing trials trial 128, frequency 15 (30.00 Hz), 1 tapers the call to "ft_freqanalysis" took 73 seconds
tfr_low_right = ft_freqanalysis(cfg, data_right);
the input is raw data with 204 channels and 125 trials the call to "ft_selectdata" took 0 seconds Default cfg.pad='maxperlen' can run slowly. Consider using cfg.pad='nextpow2' for more efficient FFT computation. processing trials trial 125, frequency 15 (30.00 Hz), 1 tapers the call to "ft_freqanalysis" took 86 seconds
The TFRs of power for the 204 planar gradiometers have now been produced. Recall that the Elekta/MEGIN system has 306 sensor locations. Each location has two orthogonal planar gradiometers.
The next step is to combine the power of two gradiometers by summing those:
cfg = [];
cfg.method = 'sum';
tfr_low_left_c = ft_combineplanar(cfg,tfr_low_left);
the input is freq data with 204 channels, 15 frequencybins and 57 timebins the call to "ft_combineplanar" took 0 seconds
tfr_low_right_c = ft_combineplanar(cfg,tfr_low_right);
the input is freq data with 204 channels, 15 frequencybins and 57 timebins the call to "ft_combineplanar" took 0 seconds
This results in 102 TFRs of each of the conditions.

Plotting the TFR of the low frequency results

In the following analysis we will focus on the time following the target onset and 'attend left' trials only. Lets first consider the TFR plot of a single channel (MEG2112):
cfg = [];
cfg.baseline = 'no';
cfg.showlabels = 'no';
cfg.xlim = [-0.5 1];
cfg.zlim = [-5e-24 5e-24] ;
cfg.channel = 'MEG2112';
ft_singleplotTFR(cfg, tfr_low_left);
the call to "ft_selectdata" took 0 seconds
the call to "ft_singleplotTFR" took 1 seconds
The power spans from -5e-24 to 5e-24 . The units are in (T/m)2/Hz. To convert to (fT/m)2/Hz multiply by 1026, i.e. the range is in 0 to 1500 (fT/m)2. In the plot note the ongoing alpha activity at ~10 Hz prior to the onset of the gratings. In response to the cue and stimulus onset there is ~4 to 6 Hz response most likely reflecting the spectral content of the event-related field.
To observe the change in power we apply relative baseline (-0.5 to -0.25 s; i.e. just prior to the target onset) (powerpost/powerpre). Note ~100% decrease of power in the alpha band after the grating onset (time 0s) until the end of the analysed window. In addition note the short-lived over 100% response in the theta band at the grating onset. This respons in the theta band likely reflecting the spectral representation of the event-related field rather than induced oscillations.
cfg = [];
cfg.baseline = [-0.5 -0.25];
cfg.baselinetype = 'relative';
cfg.showlabels = 'no';
cfg.xlim = [-0.5 1];
cfg.zlim = [0.4 1.7] ;
cfg.channel = 'MEG2112';
ft_singleplotTFR(cfg, tfr_low_left);
the input is freq data with 204 channels, 15 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_singleplotTFR" took 1 seconds

Question 1

Plot the results using an absolute baseline. Why is a relative baseline typically used in group studies?
To show the modulation in power for all sensor arranged topographically write:
cfg = [];
cfg.baseline = [-0.5 -0.3];
cfg.baselinetype = 'relative';
cfg.xlim=[-0.5 1];
cfg.zlim = [0.4 2] ;
cfg.layout = 'neuromag306planar.lay';
ft_multiplotTFR(cfg, tfr_low_left);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 204 channels, 15 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds reading layout from file neuromag306planar.lay the call to "ft_prepare_layout" took 0 seconds
the call to "ft_multiplotTFR" took 13 seconds
title('TFR of power <30 Hz')
Plot the modulation of power of the combined gradiometers, which gives a plot that is easier to interpret.
cfg.layout = 'neuromag306cmb.lay';
ft_multiplotTFR(cfg, tfr_low_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 15 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 0 seconds
the call to "ft_multiplotTFR" took 8 seconds

Question 2

Change the time-window to 0.25 s and plot the results. What is the main effect in terms of time and frequency smoothing?

Question 3

Change the time-window to 1.00 s and plot the results. What is the main effect in terms of time and frequency smoothing?
To show a topographic map of the alpha depression (10 hz) in the time interval from 350 to 1000 ms use the ft_topoplotTFR function.
cfg = [];
cfg.baseline = [-0.5 -0.3];
cfg.baselinetype = 'relative';
cfg.xlim=[0.35 1];
cfg.ylim = [10 10];
cfg.zlim = [0.5 1.5] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
figure
ft_topoplotTFR(cfg, tfr_low_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
Warning: The field cfg.showlabels is forbidden, it will be removed from your configuration
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 15 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 3 seconds

Question 4

Repeat the analysis with the 'attend right' trials. Are there any differences from the results of the 'attend left' trials?

Comparing attend right versus attend left

The alpha activty in the cue-stimulus interval is known to be highly modulated by spatial attention. Specifically, numerous studies have demonstrated posterior alpha power decrease in the hemisphere contralateral to the attended direction and alpha power increase in the ipsilateral hemisphere. To identify this effect we will illustrate the alpha modulation in the cue-stimulus interval (-0.5 to 0 s). This will be done by calculating the power lateralization index:
Note that this measure is calculated without performing baseline correction.
tfr_low_diff=tfr_low_right_c;
tfr_low_diff.powspctrm=(tfr_low_right_c.powspctrm-tfr_low_left_c.powspctrm)./(tfr_low_left_c.powspctrm+tfr_low_right_c.powspctrm);
To show the power lateralization index for all combined sensors write:
cfg = [];
cfg.xlim=[-0.5 0];
cfg.zlim = [-0.3 0.3] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_multiplotTFR(cfg, tfr_low_diff);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 0 seconds
the call to "ft_multiplotTFR" took 5 seconds
title('Right-Left')
Then plot the topography of the alpha repsonse at 10Hz and focus on the timewindow between the cue onset (-0.5 sec) and grating onset (0 sec). Lateralized attention shift has repeatedly been reported in the literature indexed by alpha power lateralization well after the cue just prior to the target onset.
cfg = [];
cfg.xlim = [-0.5 0];
cfg.ylim = [10 10];
cfg.zlim = [-0.15 0.15] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_topoplotTFR(cfg, tfr_low_diff);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
Warning: The field cfg.showlabels is forbidden, it will be removed from your configuration
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 3 seconds
The resulting plot indicate strong lateralization of the alpha response over a number of posterior channels when comparing the 'attend left' and 'attend right' trials.
To study the topography as a function of time write:
cfg = [];
cfg.xlim = [-0.6:0.2:0.6];
cfg.ylim = [10 10];
cfg.zlim = [-0.3 0.3] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_topoplotTFR(cfg, tfr_low_diff);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
Warning: The field cfg.showlabels is forbidden, it will be removed from your configuration
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds
the call to "ft_prepare_layout" took 1 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 2 seconds
the call to "ft_prepare_layout" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 3 seconds
the call to "ft_prepare_layout" took 1 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 4 seconds
the call to "ft_prepare_layout" took 1 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 3 seconds
the call to "ft_prepare_layout" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 2 seconds
the call to "ft_prepare_layout" took 1 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 2 seconds the call to "ft_topoplotTFR" took 20 seconds

Time-frequency analysis of faster frequencies (>30 hz)

For higher frequencies, a multitaper approach is typically applied. The multitapers serve to smooth the time-frequency representations in the frequency domain. The smoothing will increase with number of tapers. Since oscillatory gamma activity (30-100 Hz) is often quite broadband more spectral smoothing is required compared to lower frequencies. Several orthogonal tapers might be used for each time window prior to the calculation of the Fourier transform. The length of the time-window (cfg.t_fimwin) is ΔT = 0.25 s. It will be moved in steps of 0.05 s starting at -1.5 and ending at 1.5 (cfg.toi). The analysis will be performed from 30 to 100 Hz in steps of Δf = 2 Hz (cfg.foi) since the frequency steps are dependent on the length of the time-window ( Δf = 1/ΔT).
The number of tapers is dependent on ΔT and the required smoothing ΔFs. We chose ΔFs = 10 (cfg.tapsmofrq). The resulting tapers are defined as
K = 2*ΔT*ΔFs-1
where K is an integer required to be larger than 0.
i.e. K = 2*0.25 * 10 - 1 = 4 tapers. Note the smoothing is ±10 i.e. about 20 Hz total (see Percival and Walden, 1993). Also keep in mind that it does not make sense to reduce the smoothing to only having one taper. In that case use a Hanning taper instead.
Run the time-frequency analysis of power on the 'attend left' and 'attend right' trials separately.
cfg = [];
cfg.output = 'pow';
cfg.channel = 'MEGGRAD';
cfg.taper = 'dpss';
cfg.tapsmofrq = 10;
cfg.method = 'mtmconvol';
cfg.foi = 30:2:100;
numfoi = length(cfg.foi);
cfg.t_ftimwin = ones(length(cfg.foi),1).* 0.25;
cfg.toi = [-1.8 : 0.05: 1];
cfg.keeptrials = 'no';
tfr_high_left = ft_freqanalysis(cfg, data_left);
the input is raw data with 204 channels and 128 trials
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds Default cfg.pad='maxperlen' can run slowly. Consider using cfg.pad='nextpow2' for more efficient FFT computation. processing trials trial 128, frequency 36 (100.00 Hz), 4 tapers the call to "ft_freqanalysis" took 707 seconds
tfr_high_right = ft_freqanalysis(cfg, data_right);
the input is raw data with 204 channels and 125 trials the call to "ft_selectdata" took 0 seconds Default cfg.pad='maxperlen' can run slowly. Consider using cfg.pad='nextpow2' for more efficient FFT computation. processing trials trial 125, frequency 36 (100.00 Hz), 4 tapers the call to "ft_freqanalysis" took 654 seconds
Combine the power of two orthogonal gradiometers at the same location:
cfg = [];
cfg.method = 'sum';
tfr_high_left_c = ft_combineplanar(cfg,tfr_high_left);
the input is freq data with 204 channels, 36 frequencybins and 57 timebins the call to "ft_combineplanar" took 0 seconds
tfr_high_right_c = ft_combineplanar(cfg,tfr_high_right);
the input is freq data with 204 channels, 36 frequencybins and 57 timebins the call to "ft_combineplanar" took 0 seconds

Plotting the TFR of the high frequency results

The outcome of the TFR analysis of the 'attend left' trials will now be plotted on a single senor (MEG1923) with respect to a relative baseline (post-power / pre-power). The baseline interval is -0.5 to -0.3 s.
cfg = [];
cfg.baseline = [-0.5 -0.25];
cfg.baselinetype = 'relative';
cfg.showlabels = 'no';
cfg.xlim = [-0.5 1];
cfg.zlim = [0.7 1.3] ;
cfg.channel = 'MEG1923';
ft_singleplotTFR(cfg, tfr_high_left);
the input is freq data with 204 channels, 36 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_singleplotTFR" took 2 seconds
Note the increase in 80-90 Hz gamma power in response to the onset (t = 0 s) of the gratings.
Plot the combined gradiometers (MEG1922+1923).
cfg = [];
cfg.baseline = [-0.5 -0.25];
cfg.baselinetype = 'relative';
cfg.showlabels = 'no';
cfg.xlim = [-0.5 1];
cfg.zlim = [0.6 1.4] ;
cfg.channel = 'MEG1922+1923';
ft_singleplotTFR(cfg, tfr_high_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 36 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_singleplotTFR" took 1 seconds
To show the TFRs for all combined sensors write:
cfg = [];
cfg.baseline = [-0.5 -0.125];
cfg.baselinetype = 'relative';
cfg.showlabels = 'no';
cfg.xlim=[-0.5 1];
cfg.zlim = [0.6 1.4] ;
cfg.layout = 'neuromag306cmb.lay';
ft_multiplotTFR(cfg, tfr_high_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 36 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
the call to "ft_multiplotTFR" took 11 seconds
title('TFR of power >30 Hz')
The time-frequency response of the induced power. A relative baseline (-0.5 to -0.125 s) has been applied. Note the induced gamma band response over a couple of posterior sensors.
For the topographic representation of gamma activity between 60 Hz and 90 Hz in response to the gratings write:
cfg = [];
cfg.baseline = [-0.5 -0.125];
cfg.baselinetype = 'relative';
cfg.xlim = [0 0.6];
cfg.ylim = [60 90];
cfg.zlim = [0.8 1.2] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_topoplotTFR(cfg, tfr_high_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
Warning: The field cfg.showlabels is forbidden, it will be removed from your configuration
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 36 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 2 seconds
Here appreciate the posterior distribution of the gamma band activity.
Next, calculate the power lateralization index (using the same equations as for the lower frequencies).
tfr_high_diff=tfr_high_right_c;
tfr_high_diff.powspctrm=(tfr_high_right_c.powspctrm-tfr_high_left_c.powspctrm)./(tfr_high_right_c.powspctrm+tfr_high_left_c.powspctrm);
and display for all sensors:
cfg = [];
cfg.xlim=[0 1];
cfg.zlim = [-0.2 0.2] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_multiplotTFR(cfg, tfr_high_diff);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 0 seconds
the call to "ft_multiplotTFR" took 8 seconds
title('Right - Left')
as well as the topographic plots:
cfg = [];
cfg.xlim = [0 0.6];
cfg.ylim = [60 90];
cfg.zlim = [-0.05 0.05] ;
cfg.showlabels = 'no';
cfg.layout = 'neuromag306cmb.lay';
ft_topoplotTFR(cfg, tfr_high_diff);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
reading layout from file neuromag306cmb.lay the call to "ft_prepare_layout" took 1 seconds
Warning: The field cfg.showlabels is forbidden, it will be removed from your configuration
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds
the call to "ft_topoplotTFR" took 2 seconds

Question 5

Compare the gamma lateralization to the alpha lateralization. Interpret the alpha and gamma band lateralization in relation to neuronal processing and functional inhibition.

Question 6

Change the smoothing to 20 Hz. How many tapers does this result in? What are the consequences for the time-frequency representation?

Question 7

Perform the time-frequency analysis with a Hannning taper for the gamma band first with the 'attend left' trials:
cfg = [];
cfg.output = 'pow';
cfg.channel = 'MEGGRAD';
cfg.taper = 'hanning';
cfg.method = 'mtmconvol';
cfg.foi = 30:4:100;
cfg.t_ftimwin = ones(length(cfg.foi),1).* 0.25;
cfg.toi = [-1.8 : 0.05: 1];
cfg.keeptrials = 'no';
tfr_han_left = ft_freqanalysis(cfg, data_left);
the input is raw data with 204 channels and 128 trials
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the call to "ft_selectdata" took 0 seconds Default cfg.pad='maxperlen' can run slowly. Consider using cfg.pad='nextpow2' for more efficient FFT computation. processing trials trial 128, frequency 18 (98.00 Hz), 1 tapers the call to "ft_freqanalysis" took 95 seconds
cfg = [];
cfg.method = 'sum';
tfr_han_left_c = ft_combineplanar(cfg,tfr_han_left);
the input is freq data with 204 channels, 18 frequencybins and 57 timebins the call to "ft_combineplanar" took 0 seconds
cfg = [];
cfg.baseline = [-0.5 -0.125];
cfg.baselinetype = 'relative';
cfg.xlim=[-0.5 1];
cfg.zlim = [0.6 1.4] ;
cfg.showlabels = 'no';
cfg.channel = 'MEG1922+1923';
ft_singleplotTFR(cfg, tfr_han_left_c);
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
Warning: assuming that the default scaling should be amplitude/distance rather than amplitude
the input is freq data with 102 channels, 18 frequencybins and 57 timebins the call to "ft_freqbaseline" took 0 seconds the call to "ft_selectdata" took 0 seconds
the call to "ft_singleplotTFR" took 1 seconds

Question 8

Do the same for the 'attend right' trials.
In which way does the multitapers improves the estimate and why?

Question 9

Calculate and plot the difference between the 'attend left' and 'attend right' at 30Hz to 100Hz. Show the plots. Is there any difference between the two conditions in the gamma band?

Publications and preregistration

It is essential to report all the parameters in the frequency analysis both when submitting a preregistration and a manuscript. Note that the analysis based on a standard spectral analysis and should be independent on the toolbox applied.
Example for publications text: "The time-frequency representations were calculated using a sliding time-window approach applied to the single trials. For frequencies lower than 30 Hz we used a 500 ms time-window moved in steps of 50 ms. A Hanning window was multiplied to each time-window prior to calculating the discrete Fourier transform. This resulted in ±3 Hz spectral smoothing. The power was than estimated as the modulus square of the Fourier transform. This was done for 1 to 30 Hz in steps of 2 Hz.
For frequencies above 30 Hz a multi-taper method approach was used involving a sliding time-window of 0.25 s. To ensure sufficient spectral smoothing to the detect the gamma activity, a set of tapers derived from discrete prolate spheroidal (DPSS), or Slepian sequences, were multiplied to the time-windowed data. We used 4 tapers resulting in a spectral smoothing of ~±10 Hz. Following a discrete Fourier transform the power estimate for the respective tapers were averaged. This was done for 30 to 100 Hz in steps of 2 Hz.
The power-estimated per trial were averaged. The relative change in power was then considered with respect to a baseline of -0.5 to -0.25 ms. "

References